Reinforcement Learning — From Intuition to Algorithms
A narrative-first walkthrough of reinforcement learning, starting with everyday intuition and ending with the math behind Q-learning and DQN.
Building intelligent systems with clarity.
I'm a Senior Machine Learning Engineer at Adobe with more than a decade of experience designing algorithms, training ML models, and shipping large-scale systems for autonomous vehicles and generative AI. I hold a Ph.D. in Computer Science from The Ohio State University and have authored award-winning research across sensing, perception, and behavior prediction.
Always learning
Journey at a glance
A quick tour through the roles, research labs, and collaborations that shaped my path in AI and autonomous systems.
- 2025 — Present
Senior Machine Learning Engineer
Adobe · Creative Cloud & Firefly · San Jose, CA
Leading large-scale data pipelines, training infrastructure, and responsible generative AI initiatives that power Firefly and Creative Cloud surfaces.
- 2022 — 2024
Software Engineer, Prediction
Zoox · Bay Area, CA
Shipped multi-agent prediction models for L3/L4/L5 robotaxi fleets and co-designed the training framework and dataloaders that kept the stack fed with fresh data.
- 2018 — 2022
Senior Systems Engineer
Qualcomm Research · San Diego, CA
Led prediction for Qualcomm’s L3 highway autonomous driving stack—owning forecasting models, simulation harnesses, and post-drive analytics. Earlier built integration and test automation for the stack.
- 2018
Graduated from The Ohio State University
Ph.D. Computer Science & Engineering
Dissertation on collaborative perception and behavior prediction for intelligent transportation systems.
- 2016
Research Intern
Microsoft Research · Bangalore, India
Designed AutoCalib—large-scale traffic camera calibration with <10% speed error—in Microsoft’s video analytics platform.
- 2013 — 2018
Graduate Research Assistant
The Ohio State University · Columbus, OH
Built SmartDashCam, Soft-Swipe, RoadView, and RoadMap; taught introductory programming; collaborated with Honda on live calibration and lane-level localization.
- 2012 — 2013
Software Analyst
Standard Chartered Bank · Chennai, India
Developed reporting systems and automation scripts for global private banking infrastructure.
- 2011
Project Intern
Tata Elxsi · Chennai, India
Optimized LTE PDCCH blind decoding algorithms and explored DSP-based radio prototyping.
- 2008 — 2012
B.Tech Electrical Engineering & Minor in Mathematics
Indian Institute of Technology Madras · Chennai, India
Graduated with honors; led hostel council committees and secured top-1% rankings in national olympiads.
Featured work
A curated selection of recent publications and projects that explore robust perception, generative modeling, and multi-agent systems at scale.
-
-
The Custom Kernel Craze — Why Developers Are Taking the Wheel on GPU Optimization
Why modern AI teams are handcrafting GPU kernels—from FlashAttention to TPU Pallas code—and how smarter tooling is making silicon-level tuning accessible.
-
When AI Sees and Speaks — The Rise of Vision-Language Models
UpdatedA high level view on how modern vision-language models connect pixels and prose, from CLIP and BLIP to Flamingo, MiniGPT-4, Kosmos, and Gemini.
-
vLLM and the Trilogy of Modern LLM Scaling
How PagedAttention, Continuous Batching, Speculative Decoding, and Quantization unlock lightning-fast, reliable large language model serving.
-
Diffusion — From Molecules to Machines
UpdatedA clear introduction to diffusion and guided diffusion — how a simple physical process became a foundation for modern generative AI, from Stable Diffusion to robotics and protein design.
-
The Hidden Engine of AI — Training Frameworks and Resilience
A reader-friendly guide to scaling AI models beyond the data pipeline—from training loops and distributed frameworks to checkpoints, mixed precision, and fault tolerance.
-
The Hidden Engine of AI — Datasets and Dataloaders
A deep dive into how datasets and dataloaders power modern AI—from the quiet pipeline that feeds models to the sophisticated tools that make training efficient. Understanding the hidden engine that keeps AI systems running.
-
XGBoost — The Art of Boosting Trees with Second-Order Gradients
A deep dive into XGBoost — how second-order Taylor approximations and sophisticated regularization make it the dominant algorithm for structured data, bridging mathematical rigor with system engineering excellence.
-
The Transformer — How Machines Pay Attention
An intuitive introduction to the Transformer architecture — from the attention mechanism to self-attention and cross-attention, using language translation as a concrete example.
-
Variational Autoencoders — From Compression to Creation
An intuitive introduction to Variational Autoencoders — how compressing data into probabilistic codes enables machines to generate realistic images, sounds, and structures.
-
The Role of Predictions in Closed-Loop Autonomous Driving
Reflections on building production-grade behavior prediction systems at Zoox and Qualcomm — and why closed-loop reasoning is the bridge between perception and planning.
-
AutoCalib – Automatic Traffic Camera Calibration at Scale
UpdatedHow we used deep learning to automatically calibrate traffic cameras by observing vehicle motion—work that won Best Paper Award at ACM BuildSys 2017.
-
From Bits to Sight - A Journey from Wireless to Autonomous Vision
UpdatedMy research journey from wireless communication foundations to solving the camera calibration bottleneck that enables autonomous vehicle vision.
Recent writing
View all posts-
Tongue Twister Practice Sheet — Clarity and Fluency for Tech Talks
A structured articulation and pacing warm-up designed to help technologists speak with clarity and confidence in high-stakes meetings.
-
The 45-Minute Interview Flow: A Proven Framework for Cracking Google-Style Coding Interviews
A collaborative 45-minute thinking algorithm tuned for Google-style coding interviews—classify the problem, co-design an optimal approach, code with confidence, and handle follow-ups with ease.